compaction

1 files changed, 82 insertions, 30 deletions

diff --git a/libpanto/src/session.zig b/libpanto/src/session.zig
index c1fd1c4..21193c8 100644
--- a/libpanto/src/session.zig
+++ b/libpanto/src/session.zig
@@ -129,36 +129,10 @@ pub const DiskMessage = struct {
 
 /// Token usage reported by a provider for a single assistant turn.
 ///
-/// All four input categories sum to the total prompt tokens that were
-/// billed for the turn:
-///   `input + cache_read + cache_write` = total prompt tokens.
-///
-/// `reasoning` is a **subset** of `output`, not additive — it's the
-/// portion of the output that the model spent on internal reasoning
-/// (OpenAI o-series, Anthropic extended thinking). Cost is computed
-/// from `output` only; `reasoning` is tracked for display.
-///
-/// All fields are `u64` and default to 0. Providers that don't report a
-/// given category leave it at 0. Future cache-tier breakdowns (e.g.
-/// Anthropic's 5m vs 1h tiers) can be added as new fields without
-/// breaking the format.
-pub const Usage = struct {
-    /// Fresh input tokens billed at the model's base input rate.
-    input: u64 = 0,
-    /// Output tokens billed at the model's base output rate.
-    output: u64 = 0,
-    /// Input tokens served from cache (typ. 0.1× base input rate on
-    /// Anthropic, 0.5× on OpenAI).
-    cache_read: u64 = 0,
-    /// Input tokens written to a new cache entry (Anthropic only at
-    /// time of writing; typ. 1.25× base input rate). OpenAI's cache
-    /// hits don't bill a write premium, so this stays 0 there.
-    cache_write: u64 = 0,
-    /// Subset of `output` spent on reasoning. For OpenAI o-series
-    /// models and Anthropic extended thinking. Display-only — cost is
-    /// already accounted for via `output`.
-    reasoning: u64 = 0,
-};
+/// Defined in `conversation.zig` (so in-memory `Message`s can carry it
+/// without a module cycle) and re-exported here for the on-disk types and
+/// historical call sites that import it as `session.Usage`.
+pub const Usage = conversation.Usage;
 
 // =============================================================================
 // Content blocks
@@ -169,6 +143,7 @@ pub const DiskContentBlock = union(enum) {
     thinking: DiskThinkingBlock,
     tool_use: DiskToolUseBlock,
     tool_result: DiskToolResultBlock,
+    compaction_summary: DiskCompactionSummaryBlock,
 
     pub fn deinit(self: DiskContentBlock, alloc: Allocator) void {
         switch (self) {
@@ -176,6 +151,7 @@ pub const DiskContentBlock = union(enum) {
             .thinking => |b| b.deinit(alloc),
             .tool_use => |b| b.deinit(alloc),
             .tool_result => |b| b.deinit(alloc),
+            .compaction_summary => |b| b.deinit(alloc),
         }
     }
 };
@@ -219,6 +195,16 @@ pub const DiskToolResultBlock = struct {
     }
 };
 
+/// A compaction summary block: the synthetic seed text standing in for a
+/// compacted conversation prefix. Sits alone in a `user`-role message. See
+/// `conversation.CompactionSummaryBlock`.
+pub const DiskCompactionSummaryBlock = struct {
+    text: []const u8, // owned
+    pub fn deinit(self: DiskCompactionSummaryBlock, alloc: Allocator) void {
+        alloc.free(self.text);
+    }
+};
+
 // =============================================================================
 // File entry (header or entry)
 // =============================================================================
@@ -400,6 +386,14 @@ fn writeDiskBlock(s: *std.json.Stringify, block: DiskContentBlock) !void {
             try s.write(b.content);
             try s.endObject();
         },
+        .compaction_summary => |b| {
+            try s.beginObject();
+            try s.objectField("type");
+            try s.write("compactionSummary");
+            try s.objectField("text");
+            try s.write(b.text);
+            try s.endObject();
+        },
     }
 }
 
@@ -575,6 +569,9 @@ fn parseDiskBlock(allocator: Allocator, obj: std.json.ObjectMap) ParseError!Disk
         errdefer allocator.free(tuid);
         const content = try dupeStringField(allocator, obj, "content");
         return .{ .tool_result = .{ .tool_use_id = tuid, .content = content } };
+    } else if (std.mem.eql(u8, t, "compactionSummary")) {
+        const text = try dupeStringField(allocator, obj, "text");
+        return .{ .compaction_summary = .{ .text = text } };
     } else {
         return error.UnknownBlockType;
     }
@@ -643,6 +640,10 @@ pub fn contentBlockToDisk(
             const text = try allocator.dupe(u8, sb.text.items);
             return .{ .text = .{ .text = text } };
         },
+        .CompactionSummary => |cs| {
+            const text = try allocator.dupe(u8, cs.text.items);
+            return .{ .compaction_summary = .{ .text = text } };
+        },
     }
 }
 
@@ -681,6 +682,10 @@ pub fn diskContentBlockToInternal(
             const content = try conversation.textualBlockFromSlice(allocator, b.content);
             return .{ .ToolResult = .{ .tool_use_id = tuid, .content = content } };
         },
+        .compaction_summary => |b| {
+            const tb = try conversation.textualBlockFromSlice(allocator, b.text);
+            return .{ .CompactionSummary = .{ .text = tb } };
+        },
     }
 }
 
@@ -1031,3 +1036,50 @@ test "diskContentBlockToInternal: Thinking preserves signature" {
     try testing.expectEqualStrings("reasoning...", inmem.Thinking.text.items);
     try testing.expectEqualStrings("sig123", inmem.Thinking.signature.?);
 }
+
+test "compactionSummary block round-trips through serialize/parse" {
+    const a = testing.allocator;
+
+    var content = try a.alloc(DiskContentBlock, 1);
+    content[0] = .{ .compaction_summary = .{ .text = try dupe(a, "earlier history summary") } };
+
+    const entry: SessionEntry = .{ .message = .{
+        .base = .{
+            .id = try dupe(a, "cafef00d"),
+            .parent_id = null,
+            .timestamp = try dupe(a, "2026-04-25T17:40:00Z"),
+        },
+        .message = .{ .role = .user, .content = content },
+    } };
+    defer entry.deinit(a);
+
+    const line = try serializeEntry(a, entry);
+    defer a.free(line);
+    try testing.expect(std.mem.indexOf(u8, line, "\"type\":\"compactionSummary\"") != null);
+
+    var fe = try parseLine(a, line);
+    defer fe.deinit(a);
+    const got = fe.entry.message;
+    try testing.expectEqual(DiskMessageRole.user, got.message.role);
+    try testing.expectEqualStrings("earlier history summary", got.message.content[0].compaction_summary.text);
+}
+
+test "compactionSummary bridges in-memory <-> disk both directions" {
+    const a = testing.allocator;
+
+    // in-memory -> disk
+    const tb = try conversation.textualBlockFromSlice(a, "S1");
+    const block: conversation.ContentBlock = .{ .CompactionSummary = .{ .text = tb } };
+    defer {
+        var mut = block;
+        mut.deinit(a);
+    }
+    const disk = try contentBlockToDisk(a, block);
+    defer disk.deinit(a);
+    try testing.expectEqualStrings("S1", disk.compaction_summary.text);
+
+    // disk -> in-memory
+    var inmem = try diskContentBlockToInternal(a, disk);
+    defer inmem.deinit(a);
+    try testing.expectEqualStrings("S1", inmem.CompactionSummary.text.items);
+}